The B-field is always used self-consistently for both energy losses and synchrotron emission.
I have not studied the effect on the cosmic-ray lepton spectra, but I expect it will be small since the energy losses depend only on the total field, which is not very different in the various models (they mainly change details of the topology and regular vs random components). Hence the gamma skymaps will also not change much.
Note that one FJ implementation here (code from Farrar) is only the regular part. There is also a regular+random option, from the paper above. I have not tested this much.
Any experiences or feedback on this are welcome!
NB the GALPROP implementation is at http://sourceforge.net/projects/galprop